Full Autonomy All But Confirmed for 3, "We're doing the obvious thing"

[JeffK]

What I meant was that Mobileye is still ahead of everyone. So how can Nvidia (likely replacement), provide a faster track to full autonomy? This may be a good move for Tesla in the long run if Nvidia's tech can mature to suit requirements. But I can't imagine how this would get Tesla full autonomy any quicker. The move doesn't make sense for that purpose.

This is actually the easy part. Nvidia hardware can build DNNs faster. Tesla already has millions of miles of data and according to Tesla was using MobilEye for limited functionality and everything else was done by Tesla using other means.

Using the data they already have, they can build better networks and hardware and test it out is a fast feedback loop.

 

[flamingoezz]

maybe tesla is working on their own chipset..

they started with a lotus platform on the roadster before building their own. they started using other companies batteries before building gigafactory. would anyone be surprised given Elon's hints at autonomy before EyeQ4 is due to be ready, if Tesla was building their AP hardware/software 100% themselves?

What I meant was that Mobileye is still ahead of everyone. So how can Nvidia (likely replacement), provide a faster track to full autonomy? This may be a good move for Tesla in the long run if Nvidia's tech can mature to suit requirements. But I can't imagine how this would get Tesla full autonomy any quicker. The move doesn't make sense for that purpose.

 

[JeffK]

I'm not going to lie, I am anxious as to if Elon is going to clarify what's going on... or maybe he'll wait until the second reveal

 

[ecarfan]

What I meant was that Mobileye is still ahead of everyone.

...according to Mobileye. Others may disagree.

 

[zenmaster]

...according to Mobileye. Others may disagree.

If you bothered to point that out then it begs the question of why would they disagree? That's kind of the point, not that everyone thinks the same.

 

[JeffK]

If you bothered to point that out then it begs the question of why would they disagree? That's kind of the point, not that everyone thinks the same.

Consider that this demo from Bosch last year performs better that the most advanced MobilEye tech.

 

[Vitold]

If you bothered to point that out then it begs the question of why would they disagree? That's kind of the point, not that everyone thinks the same.

Isn't it obvious? Tesla AP is better than MobileyeQ3.

 

[zenmaster]

Isn't it obvious? Tesla AP is better than MobileyeQ3.

Yes, well how so would be the question and what is their development strategy. "Tesla AP" is a set of technologies undergoing constant development. What parts of 'Tesla AP" actually belong to Tesla.

 

[Vitold]

Yes, well how so would be the question and what is their development strategy. "Tesla AP" is a set of technologies undergoing constant development. What parts of 'Tesla AP" actually belong to Tesla.

Everyone else has access to MobileyeQ3/Bosh/etc and cannot beat AP because Tesla's advantage is it's software and constant development/updates.

 

[zenmaster]

Everyone else has access to MobileyeQ3/Bosh/etc and cannot beat AP because Tesla's advantage is it's software and constant development/updates.

What functionality of auto pilot is in Tesla's software?

 

[Vitold]

What functionality of auto pilot is in Tesla's software?

Everything, Mobileeye3 system is used just like one of the sensors. AP does all decision-making and learning.

 

[zenmaster]

Everything, Mobileeye3 system is used just like one of the sensors. AP does all decision-making and learning.

Obviously the control logic (decision making) is in Tesla's software. But that is not remotely "all of it"!

 

[alseTrick]

Hopefully this is a good sign for autopilot and autonomy and not a setback.

 

[zenmaster]

Elon said full autonomy (driverless anywhere) in 2018, while Mobileye says 2021 with its EyeQ5 chip. To make good on Elon's timeline, Tesla would need other tech. It will be interesting to see if the problem can be solved so soon by either party.

 

[KeRaSh]

I just hope the necessary hardware will be in the Model 3 on day one. I don't care if they need another 2 years after that to develop the software for it but I'd be a little bummed out if they start rolling out new models with upgraded hardware shortly after getting my Model 3.

 

[zenmaster]

I just hope the necessary hardware will be in the Model 3 on day one. I don't care if they need another 2 years after that to develop the software for it but I'd be a little bummed out if they start rolling out new models with upgraded hardware shortly after getting my Model 3.

I think you can count on a constant improvement in hardware necessary to support autonomous driving with consequent obsolescence. So you'd have to buy a new car to get the latest tech. The next best thing would be some kind of modular design for the components which would allow some upgradability.

But we don't really know what Musk means by "fully autonomous" for his 2018 goal and therefore what hardware requirements are necessary. His idea likely just means adding some degree of urban capability allowing for maybe full autonomy for 90% of time or distance from and back to your home parking spot.

That last 10% represents the inevitable, but much more difficult driving scenario problems to solve. 1% of that 10% could represent years of work and enable hundreds of new special environments and conditions. These extra scenarios will undoubtedly require more processing power and better sensors. You may think that 90% is good enough until new features provide much more convenience for your particular needs.

 

[JeffK]

I think you can count on a constant improvement in hardware necessary to support autonomous driving with consequent obsolescence. So you'd have to buy a new car to get the latest tech. The next best thing would be some kind of modular design for the components which would allow some upgradability.

But we don't really know what Musk means by "fully autonomous" for his 2018 goal and therefore what hardware requirements are necessary. His idea likely just means adding some degree of urban capability allowing for maybe full autonomy for 90% of time or distance from and back to your home parking spot.

That last 10% represents the inevitable, but much more difficult driving scenario problems to solve. 1% of that 10% could represent years of work and enable hundreds of new special environments and conditions. These extra scenarios will undoubtedly require more processing power and better sensors. You may think that 90% is good enough until new features provide much more convenience for your particular needs.

I don't know. I have two eyes and they haven't been upgraded since I got Lasik. Same brain... possibly degraded, over time.

Once you can make an accurate 3D map to a sufficient resolution at a reasonable distance for expected speeds then besides processing speed, I'm not sure there's much else to upgrade hardware-wise (assuming sufficient redundant systems)... Everything else is software.

 

[zenmaster]

I don't know. I have two eyes and they haven't been upgraded since I got Lasik. Same brain... possibly degraded, over time.

Once you can make an accurate 3D map to a sufficient resolution at a reasonable distance for expected speeds then besides processing speed, I'm not sure there's much else to upgrade hardware-wise (assuming sufficient redundant systems)... Everything else is software.

You're always going to need faster processing hardware and better sensors.

 

[JeffK]

You're always going to need faster processing hardware and better sensors.

What I'm saying is that I don't see why... there comes a point where the sensors will be able to provide all the necessary information and anything is useless. As for processing, once you achieve real-time or even just acceptable time (much faster than humans) then it'll allow fully autonomous driving. Once you have full autonomy then there's not really a need for better sensors or faster processing.

There will still room for improvement on camera sensor sensitivity and dynamic range for the next few years though.

 

[Topher]

Once you have full autonomy then there's not really a need for better sensors or faster processing.

Some fully autonomous humans are better drivers than other fully autonomous humans. There will always be edge cases where drivers (human or AI) can get better, or need to know more.

Thank you kindly.